Ace Hardware, an Unplanned Upload Partners Session, and Moving My Cloud Router to a New Host
A normal shift that turned into a long day once a real outage finally pushed me to make a change I'd been putting off.
Ace Hardware, then an unplanned evening session
Worked at Ace Hardware from 12:00 to 17:00. While I was there I got a text that I'd been scheduled for a session at Upload Partners from 18:00 to 20:00 that evening.
A real outage, and finally moving off a host that kept dropping
The cloud server that connects my own block of internet addresses to the rest of the internet went completely offline for about 55 minutes a couple of days ago, and dropped again briefly the next day. That's the final straw after a few too many of these — I decided it was time to move that whole setup to a different hosting provider instead of continuing to troubleshoot a host that keeps going down.
Spent today building the replacement from scratch on the new provider and setting it up ahead of time so it was ready to switch over at a moment's notice. Once it was ready, I did the actual cutover myself. Hit one real snag during the switch: my VPN follows whichever server answered last, and for a few minutes both the old and new servers claimed to be the same machine, so my home router kept reaching for the old one instead of switching over cleanly. My sites were unreachable for about fifteen minutes while I caught that, stopped the old side first, and restarted cleanly on the new one. Checked everything afterward from outside my own network to confirm sites were loading correctly through the new path.
Also backed up the entire setup — years of configuration that had only ever existed on that one server, nowhere else — to a private repository so a future failure like this one can't wipe it out for good.
Fixing dropped video during an overnight stress test
Ran an overnight stress test on a test unit of OmniBridge and found two real bugs worth fixing before they'd ever show up for a customer. First, video over SRT (the protocol OmniBridge and vMix use to send a camera feed over the network) was quietly sending about a fifth of its data twice, even though nothing was actually being lost — a timer assumed a piece was missing when the reply had simply not arrived back yet. Made it wait longer before assuming something's actually missing, and the unnecessary repeats dropped to zero. Fixed it on the test unit; it'll ship in the next release. Second, found that discovering other NDI video sources on the network silently failed on a Linux installation unless one extra piece of networking software was present — added it as a proper requirement so it's never missing again.
A small tool improvement
Also fixed a rough edge in my own credential-manager tool: a stored secret with several labeled fields (an API key with separate parts, for example) used to only let me read one specific field by name; now I can see every field on a saved secret at once, or pull out exactly the one I need.